We present an unmixing method, based on genetic algorithm-soil-vegetation-atmosphere-transfer modeling to extract subgrid information of soil and vegetation from remotely sensed soil moisture (downscaled; e.g., soil hydraulic properties, area fractions of soil-vegetation combinations, and unmixed soil moisture time series) that most land surface models use. The unmixing method was evaluated using numerical experiments comprising mixed pixels with simple and complex soil-vegetation combinations, in idealized case studies (with or without uncertainty) and under actual field conditions (Walnut Creek (WC11) field, Soil Moisture Experiment 2005, Iowa). Additional validation experiments were conducted at an airborne-remote sensing footprint (Little Washita (LW21) site, Southern Great Plains 1997 hydrology campaign, Oklahoma) using Electronically Scanning Thin Array Radiometer (ESTAR). Results of the idealized experiments suggest that the unmixing method can extract optimal or near-optimal solutions to the inverse problem under different hydrologic and climatic conditions. Errors in soil moisture data and initial and boundary conditions can compound uncertainty in the solution. The solutions generated under actual field conditions (WC11 field) were able to match soil moisture observations. Analysis showed that typical soil moisture retention curves of cataloged dominant soils in WC11 field did not match well with the measurements, but those derived from actual field-scale soil moisture inversion matched better. The unmixing method performed well in replicating soil hydraulic behavior at the ESTAR footprint. Unlike in WC11 field, the typical soil moisture retention curves of cataloged soils in LW21 field matched better with the measurements. We envisaged that the unmixing method can provide quick and easy way of extracting subgrid soil moisture variability and soil-vegetation information in a pixel.